Comparative Approach for Early Diabetes Detection with Machine Learning

Abstract

The detrimental effects of diabetes are currently affecting a sizeable section of the population worldwide, and many of these individuals are not being properly diagnosed. This could eventually lead to significant health issues like kidney failure and vision blindness. Chances of heart attacks and strokes increase by two to three times due to diabetes. Thus, this work has considered a total of 520 instances with included 17 features such as polyuria, gender, age, sudden weight loss, polydipsia, polyphagia, weakness, irritability, genital thrush, itching, vision blurring, muscle stiffness, alopecia, delayed healing, delayed healing, and obesity to classify the type of diabetes at an early stage to avoid such risk. Various Machine Learning (ML) methods can be employed to accurately classify the disease. The objective of this research is to predict diabetes with the help of a variety of machine learning (ML) methods and to identify the most efficient model with the highest accuracy. A total 8 classification algorithms are used for the performance measurement, these are Support Vector Classifier (SVC), Gaussian Naive Bayes (GNB), Random Forest (RF), Decision Tree Classifier (DTC), Logistic Regression (LR), Extra Tree Classifier (ETC), K-Nearest Neighbors (KNN), and XGBoost (XGB) because these models gave the highest accuracy for this dataset. After comparative analysis, the results present that Extra Tree Classifier (ETC) has the highest accuracy, i.e., 98.55%, and can be considered the best and efficient ML classification technique for diagnosing diabetes based on mentioned parameters.